PersistBench: When Should Long-Term Memories Be Forgotten by LLMs?

Sidharth Pulipaka, Oliver Chen, Manas Sharma, Taaha Saleem Bajwa, Vyas Raina, Ivaxi Sheth
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:100051-100126, 2026.

Abstract

Conversational assistants are increasingly integrating long-term memory with large language models (LLMs). This persistence of memories, e.g., the user is vegetarian, can enhance personalization in future conversations. However, the same persistence can also introduce safety risks that have been largely overlooked. Hence, we introduce PersistBench to measure the extent of these safety risks. We identify two long-term memory-specific risks: cross-domain leakage, where LLMs inappropriately inject context from the long-term memories; and memory-induced sycophancy, where stored long-term memories insidiously reinforce user biases. We evaluate 18 frontier and open-source LLMs on our benchmark. Our results reveal a surprisingly high failure rate across these LLMs - a median failure rate of $53%$ on cross-domain samples and $97%$ on sycophancy samples. To address this, our benchmark encourages the development of more robust and safer long-term memory usage in frontier conversational systems.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-pulipaka26a, title = {{P}ersist{B}ench: When Should Long-Term Memories Be Forgotten by {LLM}s?}, author = {Pulipaka, Sidharth and Chen, Oliver and Sharma, Manas and Bajwa, Taaha Saleem and Raina, Vyas and Sheth, Ivaxi}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {100051--100126}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/pulipaka26a/pulipaka26a.pdf}, url = {https://proceedings.mlr.press/v306/pulipaka26a.html}, abstract = {Conversational assistants are increasingly integrating long-term memory with large language models (LLMs). This persistence of memories, e.g., the user is vegetarian, can enhance personalization in future conversations. However, the same persistence can also introduce safety risks that have been largely overlooked. Hence, we introduce PersistBench to measure the extent of these safety risks. We identify two long-term memory-specific risks: cross-domain leakage, where LLMs inappropriately inject context from the long-term memories; and memory-induced sycophancy, where stored long-term memories insidiously reinforce user biases. We evaluate 18 frontier and open-source LLMs on our benchmark. Our results reveal a surprisingly high failure rate across these LLMs - a median failure rate of $53%$ on cross-domain samples and $97%$ on sycophancy samples. To address this, our benchmark encourages the development of more robust and safer long-term memory usage in frontier conversational systems.} }
Endnote
%0 Conference Paper %T PersistBench: When Should Long-Term Memories Be Forgotten by LLMs? %A Sidharth Pulipaka %A Oliver Chen %A Manas Sharma %A Taaha Saleem Bajwa %A Vyas Raina %A Ivaxi Sheth %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-pulipaka26a %I PMLR %P 100051--100126 %U https://proceedings.mlr.press/v306/pulipaka26a.html %V 306 %X Conversational assistants are increasingly integrating long-term memory with large language models (LLMs). This persistence of memories, e.g., the user is vegetarian, can enhance personalization in future conversations. However, the same persistence can also introduce safety risks that have been largely overlooked. Hence, we introduce PersistBench to measure the extent of these safety risks. We identify two long-term memory-specific risks: cross-domain leakage, where LLMs inappropriately inject context from the long-term memories; and memory-induced sycophancy, where stored long-term memories insidiously reinforce user biases. We evaluate 18 frontier and open-source LLMs on our benchmark. Our results reveal a surprisingly high failure rate across these LLMs - a median failure rate of $53%$ on cross-domain samples and $97%$ on sycophancy samples. To address this, our benchmark encourages the development of more robust and safer long-term memory usage in frontier conversational systems.
APA
Pulipaka, S., Chen, O., Sharma, M., Bajwa, T.S., Raina, V. & Sheth, I.. (2026). PersistBench: When Should Long-Term Memories Be Forgotten by LLMs?. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:100051-100126 Available from https://proceedings.mlr.press/v306/pulipaka26a.html.

Related Material